vol01ch01 - Over 50 Years of Image Moments and Moment Invariants

Transcription

CHAPTER
1
Over 50 Years of Image Moments and Moment Invariants
George A. Papakostas
This chapter aims to analyze the research eld of moments and moment invariants
in a holistic way. Initially, a literature analysis of the last 50 years is presented and
discussed in order to highlight the potential of this topic and the increasing interest
in many disciplines. A more in depth study of the issues addressed through the years
by the researchers is next presented both in theory and applications of moments. The
most representative works in each research direction are discussed in a chronological
order to point out the progress in each specic eld of action. This analysis concludes
with the challenges and perspectives that should motivate researchers towards the
promotion of the moments and their invariants to new scientic horizons.
For the rst time, this chapter gives a global overview of what happened in the last
50 years in moments and moment invariants research eld, but most of all it brings
to light the open issues that should be addressed and highlights the rising topics that
will occupy the scientists in the coming years. This chapter serves as a guide to those
who nd the eld of moments and moment invariants a brilliant eld of action since
it encloses all the milestones of this eld.
George A. Papakostas
Department of Computer and Informatics Engineering
Eastern Macedonia and Thrace Institute of Technology (EMaTTech)
65404 Kavala, Greece
e-mail: [email protected]
Editor: G.A. Papakostas, Moments and Moment Invariants - Theory and Applications
c Science Gate Publishing 2014
DOI: 10.15579/gcsr.vol1.ch1, GCSR Vol. 1, 3
4
G.A. Papakostas
1.1 Introduction
The rst introduction of 2-D moments in engineering life was performed by Hu in 1962
[60]. Hu proposed the 2-D geometric moments of a distribution function (an image)
as a structured element of what he called moment invariants.
In that work, Hu
used the theory of algebraic invariants in order to dene seven orthogonal invariants
to linear transformations (translation, rotation, scaling, skew).
Since then, after more than 50 years of research, a lot of new achievements in the
theory of moments and moment invariants have been presented.
The resulted new
theoretical framework has boosted the applicability of moments in many disciplines,
while a continuously increasing number of scientists have set the moments in the
center of their research.
The next milestone was the introduction of orthogonal moments by Teague [150] in
1980. Teague proposed Zernike and Legendre orthogonal moments in image analysis
as a solution to the inherent drawback of geometric moments and Hu's invariants
too, the high information redundancy. Geometric moments are the projection of the
intensity function of an image onto specic monomials, which do not construct an
orthogonal basis.
Orthogonal moments came to overcome this disadvantage of the
conventional moments since their kernels are orthogonal polynomials. The property of
orthogonality gives to the corresponding moments the feature of minimum information
redundancy, meaning that dierent moment orders describe dierent part of the image.
The rst detailed analysis of moments' properties and performance in image analysis
was performed by Teh and Chin's [151]. This analysis had inspired all the later works
in orthogonal moments and helped to understand the power of describing an image in
terms of an orthogonal polynomials' base. As a consequence of Teh and Chin [151]
work was the introduction of Zernike moments in pattern recognition by Khotanzad
and Hong [71]. Since then Zernike moments constituted the most popular moment
family due to their inherent property of being invariant under rotation and ipping of
the image.
Belkasim et al. [8] investigated the performance of the moment invariants by comparing the algebraic and orthogonal moment invariants proposed until then, in pattern
recognition applications.
This comparative study resulted to a new set of Zernike
moment invariants derived by proper normalization of the corresponding moments in
order to reduce their range.
The next milestone was the introduction of ane moment invariants by Flusser and
Suk [33].
They extended Hu's algebraic moment invariants to general ane trans-
formations by proposing four ane moment invariants.
The signicance of ane
moment invariants was their capability to recognize ane-deformed patterns commonly occurring in real pattern recognition problems such as character recognition
and shape classication.
The following 20 years moments and moment invariants have attracted the attention
of scientists towards the construction of new moment families [47, 49, 50, 43, 96,
121, 182, 186, 201, 199], the improvement of the computation accuracy [83, 95,
108, 52, 168], the development of fast computation algorithms [97, 21, 116, 53], the
embodiment of invariant properties to the moment functions[7, 16, 22, 45, 69], etc.
On the other hand an increased number of applications have been shown suitable
5
for applying moments and moment invariants such as image analysis [203, 202, 20],
pattern recognition [1, 44, 94, 103, 105, 111, 66], multimedia watermarking [3, 72,
177, 165, 158], image retrieval [170, 136, 137], medical image analysis [24, 87, 173],
forensics [89, 38, 86, 131], etc.
This chapter aims to provide an overview of the research in the eld of moments
and moment invariants since the rst introduction of moment invariants by Hu [60].
Initially, a literature analysis of the last 50 years is presented and discussed in order to
highlight the high potential of this topic. Moreover, this chapter provides an in depth
study of the issues that have been addressed through the years by the researchers both
in theory and applications of moments. The most representative works in each research
direction are discussed in a chronological order to point out the progress in each specic
eld of action. This analysis concludes with the challenges and perspectives that should
motivate the researchers towards the promotion of the moments and their invariants
to new scientic horizons. This chapter aims to serve as a guide to those who are
interested in knowing the evolution of moments and moment invariants and to promote
the current scientic achievements to the next levels.
The chapter is organized as follows: Section 1.2 briey discusses the background
of moments' denitions by providing the necessary information regarding the moment
types and their corresponding properties.
Section 1.3 presents a detailed literature
analysis by discussing the publication activity in the eld of moments and moment
invariants during the last 50 years. Section 1.4 summarizes all the attempts towards the
dissemination of knowledge about the moments such as books and organized events.
Section 1.5 reviews the directions towards theory and applications that scientists have
focused their research on, by discussing the reasons which generated each need, the
current state and the future challenges of each issue. Section 1.6 determines the big
challenges for the community of moments and nally Section 1.7 concludes the overall
chapter by highlighting the most important discussed issues.
1.2 Background
Before proceeding with the demonstration and discussion of the research directions in
moments and moment invariants, it is constructive to give a short introduction to the
fundamentals of moment functions.
Traditionally, the orthogonal image moments are considered as statistical quantities
that describe the pixels distribution inside an image's space.
Mathematically, they
are computed as the projections of an image to the orthogonal basis of the used
polynomials. From an engineering and computer science point of view, the orthogonal
moments represent the similarity between the image and a number of image patterns
formed by the kernel function of each moment family [101].
1.2.1 Moment Functions Taxonomy
The most straightforward way to classify the moment functions is based on their
dimension (number of variables). Thus, there are 1-D , 2-D and 3-D moment functions
applied on signals (one dimensional), images and volumes respectively, as illustrated
6
G.A. Papakostas
Figure 1.1: Moment functions taxonomy:
Case 1 - Dimension of the distribution
function.
in Fig.(1.2.1).
Although the moment functions can be of any number of variables, herein we are
interested in the 2-D moments of an image.
Moreover, even though there are or-
thogonal and non-orthogonal moment types the following presentation is restricted to
orthogonal moments due to their popularity.
The moment functions are characterized by the type of the polynomials base, resulting to a number of dierent moment types with specic properties.
However,
there are several other more general characteristics that can dene the moments taxonomy. Moments are classied to continuous and discrete, whereas their coordinate
space is the continuous real space or the the discrete space of the image. This case
of moments classication, regarding the type of the coordinate space along with some
representative moment families of each type, is depicted in Fig.(1.2).
Since we are
interested in the moments of an image intensity function the continuous moments
should be transformed (zero-th order approximation) to a form suitable for applying
on the image.
(p + q)-th order of any moment
N × N pixels size is dened as:
The general computation form of the
an image intensity function
f (x, y)
Mpq = N F ×
of
N
−1 N
−1
X
X
type and of
Kernelpq (x, y) f (x, y) ,
x=0 y=0
where
Kernelpq (·)
corresponds to the moment's kernel consisting of the product of
the specic polynomials [116, 101] of order
basis and
NF
p and m ,
which constitute the orthogonal
is a normalization factor. The type of Kernel's polynomial gives the
name to the moment family by resulting to a wide range of moment types Fig.(1.2).
Recently, there is an increased interest in the computation of color images [17, 69,
68, 15], by making the already proposed methods for the gray-scale images inappropriate for being applied. Therefore, the moments can be categorized according to the
depth of the intensity function 1.3 to 8-bit (gray-scale) and 24-bit (color) moment
functions.
Figure 1.2: Moment functions taxonomy: Case 2 - Coordinate space.
Figure 1.3: Moment functions taxonomy: Case 3 - Intensity depth.
7
8
G.A. Papakostas
1.2.2 Moment Invariants
Although moments are ecient descriptors of the image's content, they are sensitive
to several geometric (rotation, translation, scaling, ane) and non-geometric (blur)
transformations of the image. In order to alleviate this shortcoming scientists have proposed their invariants called
moment invariants.
These values have the same properties
as the corresponding moments and they are robust to several image deformations. The
moment invariants [8] are widely used as discrimination features in pattern recognition
and classication applications.
1.3 Literature Analysis
As it is stated in the introduction section, the main goals of this chapter is the justication of the continuously increased scientic interest in moments and moment
invariants, the presentation of the research directions showing considerable activity
and the declaration of the research actions that should be followed in order to further
improve and spread the multi-discipline utilization of moments and their invariants.
While the last two objectives will be satised via a briey description of what has been
done and what needs to be done in the eld of moments, the former objective can be
achieved by analyzing the publications related to moments in the literature through
the years.
The analysis of the literature in order to nd and count the number and type of
publications during a certain period constitutes a laborious task.
study it is decided to make use of the well known
Scopus
However, for this
bibliographic database [25],
which is commonly accepted by the scientic community and includes enough information for our analysis. The searching has been performed by applying the keywords
moments, image,
and
moment invariants,
which are necessary to appear in the Title,
Abstract, and Keywords sections of the publications.
The period of our analysis was set from the introduction of moment invariants by
Hu [60] in 1962 to the current year 2014, although the publication activity of latter
year is still in progress. Moreover, we are interested only in three types of publications
namely, Journals, Book Chapters and Conference papers. In the hereafter results the
rst two types are tackled as a single one since the number of book chapters is quite
small.
Figure 1.4, illustrates the number of papers published in the last 50 years, where
the period 1962-1980 is merged due to the very small number of published studies, in
time step of 5 years. From this plot, the upward trend of the interest in moments and
moment invariants is obvious with the 5-year period 2005-2010 being characterized by
the rapid increase of all types of publications. Moreover, the importance of moments'
eld is justied by the publication of more journals and book chapters, known for the
more rigorous review process, than conferences.
By focusing our analysis on the time period of the last 10 years it can be also
derived that 2009 was the most productive year in the history of moments and moment
invariants, during which 1,231 papers of all types have been published. This number
is very big considering the high competitiveness taking place in the eld of image
description and reects the amplication of the engagement of new scientists with the
9
7000
6000
5000
Journals & Book
Chapters
4000
Conferences
3000
Total
2000
1000
0
<1980 1985 1990 1995 2000 2005 2010 2014
Figure 1.4: Moments related publications for the past 50 years (per 5 years).
1400
1200
1000
800
Journals & Book
Chapters
600
Conferences
400
Total
200
0
Figure 1.5: Per year moments related publications for the last 10 years.
10
G.A. Papakostas
moments related topics. In the following years the publishing activity has attained the
same levels with 2009, with the number of journals and book chapters showing an
upward trend.
Conclusively, can be stated that the research in the eld of moments experiences
its highest evolution so far. The outcomes of this study should be translated to more
research activities, since time and high prior-knowledge favor the discovering and developing of the next generation frameworks in both moments' theory and applications.
1.4 Knowledge Dissemination
Apart from the publications in international journals and conferences the dissemination
of knowledge regarding moments and moment invariants, some other types of actions
have been performed in the past too. Towards this direction three popular books have
been published until now, which permit the early stage researchers in this topic to
study the basics of moments and moment invariants.
The rst book was written by Mukundan and Ramakrishnan [98] in 1998 and constituted the only text for the researchers for about 10 years. This book summarizes
the main theoretical aspects of several moment functions, with emphasis to their application in image analysis. Moreover, it discusses and proves analytically some useful
properties of the moment functions and reviews the main publications proposed until
then.
The second book was written by Pawlak [120] in 2006.
This book provides a
dierent approach to the moment theory from the previous book, since it focuses
on the reconstruction performance of the moment functions, with emphasis to their
computation accuracy. The advantage of this book is that it is available [120] for free
downloading and thus anyone can retrieve it.
The third book was written by Flusser et al.
[36] in 2009, only 3 and 10 years
after the Pawlak's and Mukundan's books respectively.
The mentioned time span
is signicant since it gives us an indication of the dierences between the books'
contents and thus the amount of novel information they contain. The third book can
be compared only with the Mukundan's book, since it constitutes its update version
enriched with the theory developed in the meanwhile and with emphasis to pattern
recognition applications. These dierences make the third book the current textbook
for any researcher in this research eld.
Finally, for the dissemination of moments a special session in the International
Conference of Image Analysis and Recognition (ICIAR) was organized recently by AlRawi et al. [2]. However, the small number of contributed papers and its temporary
nature have shown that additional eorts should be made in collaboration with the
most eminent scientists in this eld in order to establish a frequent annual event
(e.g. special session or workshop) as the major meeting for the moments' knowledge
dissemination.
11
1.5 Research Directions
Although the moment functions and their invariants were introduced almost 50 years
ago, their evolution was extremely signicant only in the last 20 years. The development of novel theoretical frameworks have boosted the disciplines of their application.
In this section, it is attempted to present some important snapshots of progress in
both moments' theory and applications.
1.5.1 Theory
It is well known that the dissemination of a specic research eld is highly dependent
on the amount and accuracy of the theoretical framework that supports its scientic
correctness. Following the aforementioned trend and towards the alleviation of specic weaknesses and limitations of the fundamental moment theory, scientists have
built new tools and methods.
The most representative achievements in the theory
of moments and moment invariants are summarized and highlighted in the hereafter
subsections.
1.5.1.1 Fast Computation
Due to the fact that the computation of a moment or a moment invariant consists
of the evaluation of the moment's base in each point of the distribution function
(intensity function in the case of an image), the whole procedure is time consuming.
The computation time is further increased in the case of the moments of an image
(2-D or 3-D), since the moment's base should be evaluated across each dimension of
the intensity function. Moreover, when a set of orthogonal moments is to be computed
the computation time increases exponentially because of the high complexity of the
polynomial basis.
Therefore, the development of fast computation algorithms [116]
were a primary target of the scientists for many years, while their achievements helped
towards the computation of moments for big image data.
Mainly, the developed algorithms which ensure high computation speeds of image moments are divided into two dierent approaches: (1)
level.
Polynomial level
and 2)
Pixels
Moreover, the latter approach is realized under three possible alternatives:
Approach 1 (Polynomial level) -
The most common practice to reduce the time
needed to compute the polynomial basis is to apply a recurrence formula to compute each polynomial order by using polynomials of lower orders.
Several algo-
rithms presented in the past introduced recursive algorithms which avoid the direct
computation of the polynomials of any order, instead simplied recursive formulas
were proposed.
Such algorithms have been applied for the computation of Zernike
[74, 122, 97, 21, 104, 106], Legendre [97], Fourier-Mellin [102, 162] moments etc. It
is worth noting that some polynomials are equipped by recurrence computation formulas on their own such as Tchebichef [96], Krawtchouk [186], dual-Hahn [187, 201]
Gaussian-Hermite [182] moments etc.
12
G.A. Papakostas
Approach 2 (Pixels level) -
Another way to reduce the computation time of mo-
ments is to decrease the number of pixels where the polynomials are evaluated without loosing useful information. This can be achieved by either exploiting polynomials'
symmetry properties (symmetric pixels have identical contribution to the moment calculation) or by treating the image as a set of intensity slices consisting of homogenous
(with the same intensity value) rectangular blocks. According to the former strategy
[96, 185, 61, 199] it is not necessary to compute the polynomial values for all symmetric pixels but only for a small portion of them, depending on the symmetry type. The
latter strategy [145, 112, 115, 113, 134, 47] decomposes the moments computation
to partial computations over rectangular homogenous blocks thus their moments can
be derived easier. These two computation schemes show some limitations, the rst
strategy is applied only for those polynomials which exhibit symmetries and the second
one can be applied only for polynomials dened in the Cartesian coordinate space. In
this category can be also included an alternative computation scheme [53, 56]which is
making use of the separability property of the moment transform permitting the computation of the moments in two steps, by successive computation of the corresponding
1-D moments for each row.
It is worth noting that some of the above algorithms can operate in combination
[116, 113] under specic congurations in order to take advantage of each algorithm's
acceleration capabilities. Furthermore, although these algorithms have been applied to
compute the moments of an entire image, they can be used to compute the moments
of image's partitions in a block-based computation scheme [116]. In this case, when
image reconstruction is needed some partitioning eects occur, where each sub-image
generates dark edges that degrade the overall reconstructed image quality.
Finally, following the recent trends of applying high performance hardware and software schemes for time consuming tasks, some researchers have proposed GPU based
moments computation algorithms [153, 180, 64, 127]. Although, the introduced GPU
accelerated algorithms show an improved computation speed, there is still room for
more ecient accelerated algorithms that take advantage of the GPU resources (cores,
threads, blocks, shared memory).
1.5.1.2 Computation Accuracy
According to the moments taxonomy presented in Section 1.2 the moments functions
are classied to continuous and discrete in respect to the nature of their coordinate
space. Only for the case of continuous moment functions, their computation over a
the discrete pixels space of an image, encounters some inaccuracies. These errors are
of two types namely
geometric
and
numerical
errors [83, 168, 178]. The rst type of
errors is caused by the projection of a square discrete image onto the domain (e.g. the
unit disc for the radial polynomials) of the polynomial basis, while the numerical errors
are occur due to the calculation of the double integral over xed sampling intervals,
by applying the zeroth order approximation.transformation.
Several approaches have been proposed in the literature towards the minimization
of both error types. More precisely, the geometric errors are minimized by applying
specic mapping techniques from the image space to the polynomials [168] domain
and appropriate pixels arrangement methodologies [83].
13
The numerical integration
errors are decreased by applying either analytical [168, 52, 56] or approximate iterative
integration algorithms (e.g. Simpson, Gauss) [83, 55]. Currently, by using the aforementioned techniques the derived moment values are very close to their theoretical
values and thus the achieved accuracy level is satisfactory.
1.5.1.3 Numerical Stability
Apart from the aforementioned computation errors, caused by the inherent weaknesses
to apply the mathematical formulas to the set of image's pixels, the computation of
image moments reveals some additional numerical instabilities [95, 108, 102, 109, 142,
143, 144, 126].
Recently, the author and his colleagues have analyzed the numerical behavior of
the recursive algorithms for Zernike moments computation [108, 109].
that under certain circumstances some truncation errors called
They found
nite precision errors
occur in an iteration of the algorithms, which is increased iteration by iteration and
nally they cause the collapse of the algorithm. These errors are generated by specic
mathematical operations such as subtraction and division.
The other common numerical instability occurring during the computation of image
moments are the
overows
[102, 106] due to the existence of big numbers and the
great amount of operations between them.
The overows are more frequent with
the increase of moment orders and the size of the image.
This is the reason why
the reconstruction of an image is considered only for small images (<
1024 × 1024
pixels). Moreover, the presence of big quantities during the computation causes large
variations in the dynamic range of moment values, [95].
In order to overcome this
situation scaled moments were proposed[96, 185].
The numerical instabilities are responsible for the limitation of the maximum moment
computed order along with the limitation to the size of the image being processed.
1.5.1.4 Invariance Embodiment
The introduction of moment invariants under the basic geometric transformation
(translation, scale and rotation) by Hu [60], has shown the way to the most important
application of moment function, to pattern/object recognition and data classication
problems. The ve Hu's invariants were based on the conventional geometric moments
and their normalized versions. However, due to the description limitations (e.g. high
information redundancy) of the geometric moments and the rising of the orthogonal
moments, the moment invariants have been revised.
Mainly, there are two types of methodologies that derive invariant moments of an
image, either by image coordinates normalization and description through the geometric moment invariants [98] or by developing new computation formulas which
are characterized by invariant properties inherently [7, 22, 200]. In the rst type of
methods, we can also categorize the algebraic moment invariants since they use the
geometric moments as a structural element.
Realizing the importance of describing and recognizing a scene/object/pattern despite the position, orientation, scale etc. of the region of interest inside an image, several moment invariants have been proposed: Ane [33, 148], Rotation [31, 32, 174],
14
G.A. Papakostas
Geometric [179, 45], Orthogonal [185, 184, 69, 183], Blur [35, 193], Projective [189]
moment invariants.
Recently, combined moment invariants which ensure invariant moment description
under more than one geometric and non-geometric transformations, have been introduced [146, 197, 193, 16]. The combined invariants are very useful but they derive
dicultly. The development of invariant moments under multiple geometric and nongeometric image transformations in combination, constitutes one of the hot topics in
the eld of moments.
Moreover, it is worth noting that a particular type of moment invariants called
Complete
moment invariants [39, 194] exhibiting some very useful properties, has been
reported in the literature. The construction of a complete set of moment invariants is
performed in terms of the corresponding moments of the same orders. This description
has the advantage of enabling the inverse computation of the contributed moments
from the corresponding invariants and vice versa (duality) .
1.5.1.5 Novel Moment Families
In the last 10 years there was an increased interest in developing new moment families
and their corresponding invariants. As a result of this intense action is the introduction of new types of moments having improved properties as far as their description
capabilities and invariance behavior are concerned.
In this context, the group of Geometric [60], Zernike [150], Pseudo-Zernike [150],
Fourier-Mellin [133], Legendre [150] traditional moments, initially was enriched with
the Tchebichef [96], Krawtchouk [185], dual-Hahn [201, 187], Racah [199], discrete
moments exhibiting high computation accuracy. More moments such as Polar Harmonic Transforms [184, 49, 48] Wavelet [11, 117, 141], Gaussian-Hermite [182],
Bessel-Fourier [175], Jacobi-Fourier [121, 12], Gegenbauer [82, 55], Charlier [46], Comoments [189], Exponent [59], Variant [43] and Spline [19] moments were introduced
in a way to nd more informative and robust descriptors.
Recently, a new type of moments called
separable moments
[196, 47] was proposed.
The separable moments are constructed by using a combination of dierent polynomials for each dimension. In this way the separable Chebyshev-Gegenbauer, GegenbauerLegendre, Tchebichef-Krawtchouk, Krawtchouk-Hahn etc. [196] and Charlier-{Tchebichef,
Krawtchouk, Hahn} [47] were proposed. The separable have shown improved description capabilities compared to their non-separable versions, while their application in
several disciplines constitutes a new eld of action.
1.5.1.6 Color Moments
The increased usage of camera equipped devices in the every day life such as computers,
mobiles, tablets etc. has triggered the need for ecient processing of color images. In
the eld of moments and moment invariants little work has been reported regarding
the computation of moment function for color images.
The most straightforward practice is to compute the moments of each color channel separately [94, 147] and use them as 3-tuples for image analysis and recognition
purposes. However, this approach has the disadvantages of applying the computation
15
scheme for three times, by adding signicant time overhead and the color information
is not described in a compact way in a single number. Another similar approach is
to rst apply a color space transform (RGB to HSV) in order to isolate the color information into a single channel where the moments are being computed. In this case
non-color useful information might be discarded.
However, in the last 2-3 years the perspective of applying quaternion analysis (a
generalization of the complex analysis) in order to describe the color information of an
image in a compact way has been promoted. The rst introduced quaternion moments
were the Quaternion Fourier-Mellin moments proposed by Guo and Zhu [42] in 2011.
According to quaternion analysis, each image pixel is represented as a four-dimensional
number called quaternions. After the rst introduction of quaternion moments, the
Quaternion Zernike [17, 15] and Quaternion Bessel-Fourier [132] moments and moment
invariants [41, 93] were proposed.
The author and his colleagues have developed a
unied methodology [68, 70] to produce quaternion moments and moment invariants
of any polynomial type, by giving scientists the option to derive the most appropriate
to their applications color moments.
1.5.1.7 3D Moments and Moment Invariants
The evolution of the stereo imaging, by providing scene information in the threedimensional (3-D) coordinate space, along with the development of cheap stereo cameras, has generated the need for moments computation of volumes.
Although the
concept of 3-D moments is not new [130, 13, 100], few works for handling 3-D moments and moment invariants have been reported. This is probably due to the small
number of 3-D volumetric images that are incorporated in the every day life of humans.
However, in the last years there is an increased interest in accelerating and improving
the description capabilities of the 3-D moments, while novel 3-D moment families and
moment invariants have been proposed [149, 54, 172].
What is worth investigating, is the way the previously introduced computation algorithms for 2-D images can be applied for the case of 3-D volumetric data. This research
direction can lead to more unied computation schemes permitting the computation
of the moments of any image modality.
1.5.1.8 Moments Selection
A common practice in using moments, is the computation of all moments up to
a certain order and use them supposing that they describe adequately the image's
content.
However, this ad hoc usage of image moments is not optimal, in the
sense that no prior knowledge regarding the problem at hand is taken into account,
to guarantee the selected moments are the most appropriate for the specic task. A
possible solution to this issue is the application of an additional process that selects,
from a large pool, the moment features which best perform in terms of description
accuracy (e.g. reconstruction error, recognition rate).
To this direction, few works have been reported in the literature applying a selection
mechanism in order to select the most appropriate moments for a specic application.
The most of these methods selected moment features in a
wrapper
scheme by taking
16
G.A. Papakostas
into account the model (e.g.
classier) that handle the selected moments through
the application of an Evolutionary optimization algorithm such as a Genetic Algorithm
(GA) [107, 67, 101].
However, the main disadvantage of the GA-based selection is its high computation
time for converging to a suitable solution. This drawback makes the
methods [101] an attractive alternative approach.
lter
selection
These methods do not use the
mining model, instead the internal data properties/characteristics (dependency, correlation etc.) are taken into consideration. It is worth noting that this research direction
is open, since the need for a fast and adequate moment selection methodology which
takes into account the particularities of each application still exists.
1.5.2 Applications
From the previous sections presenting the theoretical aspects of moments and moment
invariants can be concluded that the evolution of the moment functions theory has been
also motivated by the increased needs of using moments and their invariants in many
disciplines. Although the applications of the moments and their invariants increase year
by year, there are some elds of applications where moments have provided important
solutions when compared with other similar methodologies. The applications where
the moments have shown signicant impact through the years constitute the subject
of this section.
This impact is analyzed by presenting the way moments and their
invariants are being incorporated.
1.5.2.1 Image Analysis
Although moment functions can be of any dimension, their 2-D realization has found
signicant applications in image analysis. The ability of image moments to capture
the content information of an image in a compact way and with minimum redundancy,
makes them appropriate to describe the pixels distribution of the image uniquely. Due
to these important properties image moments have been used successfully in texture
segmentation [160], image registration [34, 23], sub-pixel edge detection [40, 10],
rotation angle estimation [125, 73], image compression [99, 124], image denoising
[65, 188], shape analysis [92, 203, 202] and image matching [20].
Recently, image
moments have shown promising performance in describing the quality of images [169,
152] by giving a quality index close to human's perception.
1.5.2.2 Pattern Recognition
The main properties of image moments and their invariants are the ability to describe
uniquely the pixels distribution and the robustness to geometrical and non-geometrical
transformations of the image's content.
These two properties make the usage of
moments in pattern recognition and classication applications, where some image
patterns have to be distinguished, an important utility. The application of moment
and moment invariants in pattern recognition origins back to the early works of [60,
28, 1]. Since then, they are applied with remarkable performance on sketched symbol
[58], gait recognition [135], target recognition [84], aircraft recognition [28, 191]iris
recognition [88, 51], hand gesture recognition [123, 4], facial expression recognition
17
[75, 66], infrared face recognition [110, 29], human action classication [181], trac
sign recognition [30], texture classication [90].
1.5.2.3 Multimedia Watermarking
The introduction of moments and moment invariants in image watermarking was performed by Alghoniemy [3], who used the Hu's moment invariants and an iterative
process in order to hide the watermark into an image. Kim and Lee [72], make a step
forward by using the orthogonal Zernike moments to hide the watermark information,
in order to take advantage of the rotation invariance and reconstruction capabilities of
the radial polynomial basis. A milestone in moment-based watermarking, was the pioneering work of Xin et al. [177], which changed the way the watermark is inserted and
extracted. In this approach the dither modulation is applied by adding a blind nature
to the whole watermarking procedure since the initial watermark was not necessary
any more.
Among the several moment-based methodologies [158] incorporating moments as
information carriers, new moment families [78, 166, 139, 140, 154, 155], moments
with improved local behavior [185, 26, 118, 119], RST (Rotation, Scale, Translation)
invariance capabilities [198, 195, 166] and robustness to ane transformations [192]
have been introduced lately.
The most emerging topics in moment-based watermarking is the application of the
newly proposed quaternion moments as features to hide the watermark information
inside color images [157, 156, 167] and the tackling of watermarking as an adaptive
process [159, 119, 156, 154, 140] where the signicance of each moment coecient
and the image insertion portion are dynamically decided.
Finally, it should be noted that moments have not been used only in image watermarking but in video [161] and audio [165]watermarking as well.
1.5.2.4 Image Retrieval
Nowadays, there are massive amounts of image data due to the many camera equipped
devices (computers, mobiles, tablets, etc.) and the unstoppable usage of social media. This high trac and storage of image data assumes the existence of big image
databases, where the task of searching and retrieving images with specic attributes
constitutes an every day task for the modern information systems. Therefore, there
are needs for accurate, fast and reliable image retrieval systems. Concerning the Content Based Image Retrieval (CBIR) scheme, the usage of ecient image features to
describe the content of the images can ensure a high retrieving performance.
Image moments and moment invariants have been used successfully in CBIR image
retrieval systems for several years [170, 171, 6, 100, 136, 5]. Moments can describe
uniquely the global, as well as the local information [137, 138] of the image's content
and thus can be used to distinguish dierent images and to provide high matching
rates of the query image.
Although the application of moment functions as features in retrieving images from
a database has been already investigated, there are a lot of new moment families
that are not used for CBIR purposes.
Moreover, quaternion moments and moment
18
G.A. Papakostas
invariants [68, 70] could be promising features to retrieve color images, since they can
enclose all the color information in a single hyper complex quantity.
1.5.2.5 Medical Image Analysis
The application of moment functions in medical image analysis follows almost the same
directions with the conventional image analysis. The orthogonal moments have been
used widely to reconstruct CT images [163, 24] and noisy CT, MRI, X-ray medical
images [57], to describe the texture of a CT liver image [9], while several moment
invariants have been used as discriminative features to detect tumors [63], to predict
protein structures [176], to recognize parasites [27] and spermatogonium [87] and to
segment medical images [91, 37].
The acceptable performance of the moments in medical image analysis justies their
ability to describe complex image patterns and generates many expectations for a more
systematic incorporation of the recent advances in moments' theory to medical image
analysis.
1.5.2.6 Forensics
Image moments, as features describing uniquely the content of an image, have found
application to a quite recent topic of information security called
tion.
image forgery detec-
Moreover, the invariant properties of moments are showing very useful for the
aforementioned task, since they remain unchangeable when common geometric and
non-geometric attacks are applied on the image by any malicious user.
More precisely, the orthogonal moment invariants have been used to detect the copymove image forgery [89, 81, 79, 62] according to which a part of an image is copied and
pasted into another place inside the same image, generating duplicate regions. The
radial moment invariants can also handle successfully the copy-rotate-move [129, 128]
image forgery since they are rotation invariant and thus remain unaected to the
rotation attack.
This research eld is still unexplored regarding the application of the moments and
their invariants and thus a lot of open issues exist that can be addressed in the future
by taking advantage of the recent achievements in the moments' theory.
1.5.2.7 Miscellaneous
Apart from the above applications of moments and moments invariants, there are
several other disciplines where the moments have little contribution but the promising
results have shown that there is still room for more. Such areas are camera calibration
[77], robot localization [76], visual servoing [14], audio content authentication [86, 85],
music identication [80], spectral analysis [190, 18], strain analysis [164] and many
more.
19
1.6 New Horizons
The previous analysis of the most emerging research topics both in theory and applications of moment functions having attracted the scientic interest so far, provides a
measure of the activity taken place in the last the 50 years in the eld of moments and
moment invariants. However, the aforementioned analysis also sets the current needs
that have to be satised through future research by establishing some new horizons.
Some distinctive directions that can lead to new lands of innovation and can motivate
the early stage as well as the experienced scientists, are presented hereafter.
1.6.1 Local Behavior
Moments and moment invariants, initially were proposed as global image descriptors,
since they are computed over the entire image intensity function. This global encoding
mechanism makes them robust to noisy conditions since the lower order moments
describe the coarse image content, whereas the noise contaminates the image's details.
Moreover, the discrimination power of the moment descriptors is distributed over all
orders and thus the local information of the image is shared to many components,
by making them less ecient to describe the particular properties of a local image
region. For example, for textured images, where useful information is highly localized,
the global moments are not able to describe the high variability of the texture microstructures.
Some attempts to increase the local description capabilities of the moments have
already been reported in the literature for texture classication [90], invariant pattern
recognition [114] and image watermarking [26, 119]. However, we are still far away
from a theoretical framework that dene a local information preserving mechanism
during the computation of moment descriptors and more eorts have to be made.
1.6.2 Combined Invariants
As it has already been discussed in Section 1.5.1.4, the moment invariants constitute
a useful tool in describing the contents of an image despite the presence of some
common geometric and non-geometric deformations.
Recently, the embodiment of
multiple invariant properties [146, 197, 193, 16] to the moments have become a hot
research topic. The development of combined moment invariants to rotation, scaling,
translation, blur and ane transformations is still in the beginning and deserves the
attention of researchers.
The development of moment invariants robust to multiple geometric and nongeometric image transformations simultaneously and not partially, would show novel
directions towards the ecient and unied handling of image deformations.
1.6.3 Selection
One of the most challenging open issue in the theory of moments and moment invariants is the development of an analytical methodology for the selection of the most
appropriate moment set for a specic application and image modalities. Currently, a
20
G.A. Papakostas
set of moments up to a certain order is computed and used as image descriptors, a
practice that does not guarantee the highest performance relative to the problem's
objectives. Although, some preliminary approaches [107, 67, 101] have been proposed
in the past, as analyzed in Section 1.5.1.8, they have major shortcomings mainly due
to high computation time.
The ultimate goal towards the rising of this horizon is the development of a moments
and/or moment invariants selection scheme guaranteeing the optimality of the selected
features set subject to a specic problem. The development of such a methodology
will boost the performance in all the applications described in the previous sections.
1.6.4 Software Library
An exhaustive search in the web can lead to the conclusion that there is not any software package or library in any programming language and environment being dedicated
to moments and moment invariants. The majority of the researchers that propose their
own algorithms and tools, do not give attention in preparing their source codes in a
form suitable to distribute them to the moments community and thus the dissemination of the eld is restricted. There is a need to compile a formal and open-source
library in an advanced programming environment (MATLAB, C++, Python, R, etc.)
by incorporating the major achievements in the theory of moments.
In this way, it
is strongly believed that the evolution of the eld will be boosted, new improved algorithms will be generated and all the algorithms would be compared in the same
base.
1.7 Discussion
For the rst time, this chapter presents a comprehensive study of the research in the
eld of moments and moment invariants in the last 50 years. The literature analysis
showed the continuously increasing interest in the eld of moments but most important
that the eld's activity is at the zenith so far, a conclusion that can enforce the research
in multiple directions.
The previous sections provided an overview of the attainments achieved through the
years in each theoretical aspect of moments and where these achievements have found
application.
As an outcome of the point to point description of the past research
actions and considering the current needs of the scientic community, new horizons
of research are declared for the future.
This chapter can serve as a guide to the early stage and experienced researchers, who
are interested in focusing their research to the moments' theory and their applications.
References
[1] Y.S. Abu-Mostafa and D. Psaltis.
ants.
Recognitive aspects of moment invari-
IEEE Transactions on Pattern Analysis and Machine Intelligence,
6(6):698706, November 1984.
PAMI-
[2] M.S. Al-Rawi, K.M. Hosny, and E. Walia.
21
The methods of moments in the
analysis of images and objects. International Conference on Image Analysis and
Recognition (ICIAR), 2013. www.aimiconf.org/iciar13/specialsession-02.php.
[3] M. Alghoniemy and A.H. Tewk. Image watermarking by moment invariants. In
International Conference on Image Processing (ICIP),
volume 2, pages 7376,
10-13 Sept. 2000.
[4] G. Amayeh, G. Bebis, A. Erol, and M. Nicolescu. Peg-free hand shape verication
using high order Zernike moments.
Recognition Workshop (CVPRW),
Conference on Computer Vision and Pattern
page 40, 2006.
[5] J.P. Ananth and V.S. Bharathi. Face image retrieval using Tchebichef moments.
Indian Journal of Computer Science and Engineering,
2(6):798801, 2012.
[6] F.M. Anuar, R. Setchi, and Y.K. Lai. Trademark image retrieval using an integrated shape descriptor.
Expert Systems with Applications,
40(1):105121,
2013.
[7] S. Belkasim, E. Hassan, and T. Obeidi. Explicit invariance of Cartesian Zernike
Pattern Recognition Letters,
moments.
28(15):19691980, 2007.
[8] S.O. Belkasim, M. Shridhar, and M. Ahmadi.
Pattern recognition with mo-
ment invariants: a comparative study and new results.
Pattern Recognition,
24(12):11171138, 1991.
[9] V.S. Bharathi and L. Ganesan. Orthogonal moments based texture analysis of
CT liver images.
29(13):18681872, 2008.
[10] T.J. Bin, A. Lei, C. Jiwen, K. Wenjing, and L. Dandan. Subpixel edge location
based on orthogonal FourierMellin moments.
Image and Vision Computing,
26(4):563569, 2008.
[11] A. Broumandnia and J. Shanbehzadeh. Fast Zernike wavelet moments for Farsi
character recognition.
25(5):717726, 2007.
[12] C. Camacho-Bello, C. Toxqui-Quitl, A. Padilla-Vivanco, and J. Baez-Rojas.
High-precision and fast computation of JacobiFourier moments for image de-
J. Opt. Soc. Am.,
scription.
31(1):124134, 2014.
[13] N. Canterakis. 3D Zernike moments and Zernike ane invariants for 3D image
analysis and recognition. In
Scandinavian Conference on Image Analysis (SCIA),
pages 8593. DSAGM, 1999.
[14] F. Chaumette. Image moments: a general and useful set of features for visual
servoing.
IEEE Transactions on Robotics,
20(4):713723, 2004.
[15] B. Chen, H. Shu, G. Coatrieux, G. Chen, X. Sun, and J.L. Coatrieux.
image analysis by quaternion-type moments.
and Vision.
Color
Journal of Mathematical Imaging
in press.
[16] B. Chen, H. Shu, H. Zhang, G. Coatrieux, L. Luo, and J.L. Coatrieux. Combined invariants to similarity transformation and to blur using orthogonal Zernike
moments.
IEEE Transactions on Image Processing,
20(2):345360, 2011.
[17] B.J. Chen, H.Z. Shu, H. Zhang, G. Chen, C. Toumoulin, J.L. Dillenseger, and
L.M. Luo.
Quaternion Zernike moments and their invariants for color image
analysis and object recognition.
Signal Processing,
92(2):308318, 2012.
[18] J. Chen, B.Q. Li, H.L. Zhai, W.J. Lu, and X.Y. Zhang. A practical application
of wavelet moment method on the quantitative analysis of Shuanghuanglian oral
liquid based on three-dimensional ngerprint spectra.
Journal of Chromatogra-
22
G.A. Papakostas
phy A,
1352:5561, 2014.
[19] W. Chen, Z. Cai, and D. Qi. A new class of orthogonal Spline moments and
its application.
Journal of Information & Computational Science,
10(14):4563
4571, 2013.
[20] Z. Chen and S.K. Sun.
A Zernike moment phase-based descriptor for local
image representation and matching.
19(1):205219, 2010.
[21] C.W. Chong, P. Raveendran, and R. Mukundan.
A comparative analysis of
algorithms for fast computation of Zernike moments.
36(3):731742, 2003.
[22] C.W. Chong, P. Raveendran, and R. Mukundan. Translation and scale invariants
of Legendre moments.
[23] X. Dai and S. Khorram.
37(1):119129, 2004.
A feature-based image registration algorithm using
improved chain-code representation combined with invariant moments.
Transactions on Geoscience and Remote Sensing,
IEEE
37(5):23512362, 1999.
[24] X.B. Dai, H.Z. Shu, L.M. Luo, G.N. Han, and J.L. Coatrieux. Reconstruction of
tomographic images from limited range projections using discrete Radon transform and Tchebichef moments.
43(3):11521164, 2010.
[25] Scopus Bibliographic Database. www.scopus.com.
[26] C. Deng, X. Gao, X. Li, and D. Tao. A local Tchebichef moments-based robust
image watermarking.
Signal Processing,
89(8):15311539, 2009.
[27] E. Dogantekin, M. Yilmaz, A. Dogantekin, E. Avci, and A. Sengur. A robust
technique based on invariant moments ANFIS for recognition of human parasite eggs in microscopic images.
35(3):728
738, 2008.
[28] S.A. Dudani, K.J. Breeding, and R.B. Mcghee. Aircraft identication by moment
invariants.
IEEE Transactions on Computers,
C-26:3946, 1977.
[29] S. Farokhi, S.M. Shamsuddin, U.U. Sheikh, J. Flusser, M. Khansari, and K.J.
Khouzani. Near infrared face recognition by combining Zernike moments and
undecimated discrete wavelet transform.
Digital Signal Processing,
31:1327,
2014.
[30] H. Fleyeh, M. Dougherty, D. Aenugula, and S. Baddam.
recognition with Fuzzy ARTMAP and Zernike moments.
Vehicles Symposium (IVS),
Invariant road sign
In
pages 3136, 2007.
[31] J. Flusser. On the independence of rotation moment invariants.
nition,
IEEE Intelligent
Pattern Recog-
33(9):14051410, 2000.
[32] J. Flusser.
Recognition,
On the inverse problem of rotation moment invariants.
35(12):30153017, 2002.
[33] J. Flusser and T. Suk. Pattern recognition by ane moment invariants.
Recognition,
Pattern
26(1):167174, 1993.
[34] J. Flusser and T. Suk.
A moment-based approach to registration of images
with ane geometric distortion.
Sensing,
Pattern
IEEE Transactions on Geoscience and Remote
32(2):382387, 1994.
[35] J. Flusser, T. Suk, and S. Saic. Recognition of blurred images by the method
of moments.
IEEE Transactions on Image Processing, 5(3):533538, 1996.
Moments and Moment Invariants in Pattern
[36] J. Flusser, T. Suk, and B. Zitova.
Recognition.
23
Wiley, Chichester, United Kingdom, 2009.
[37] S.W. Franklin and S.E. Rajan. Retinal vessel segmentation employing ANN technique by Gabor and moment invariants-based features.
Applied Soft Computing,
22:94100, 2014.
[38] G. Gao and G. Jiang.
A lossless copyright authentication scheme based on
BesselFourier moment and extreme learning machine in curvature-feature domain.
Journal of Systems and Software,
86(1):222232, 2013.
[39] F. Ghorbel, S. Derrode, R. Mezhoud, T. Bannour, and S. Dhahbi. Image reconstruction from a complete set of similarity invariants extracted from complex
moments.
[40] S. Ghosal and R. Mehrotra.
detection.
27(12):13611369, 2006.
Orthogonal moment operators for subpixel edge
26(2):295306, 1993.
[41] L. Guo, M. Dai, and M. Zhu. Quaternion moment and its invariants for color
object classication.
Information Sciences,
273:132143, 2014.
[42] L.Q. Guo and M. Zhu. Quaternion FourierMellin moments for color images.
44(2):187195, 2011.
[43] J.A. Martin H., M. Santos, and J. de Lope. Orthogonal variant moments features
in image analysis.
180(6):846860, 2010.
[44] J. Haddadnia, M. Ahmadi, and K. Faez. An ecient feature extraction method
with pseudoZernike moment in RBF neural network-based human face recognition system.
EURASIP Journal on Applied Signal Processing, 2003(9):890901,
2003.
[45] M.S. Hickman. Geometric moments and their invariants.
ical Imaging and Vision,
Journal of Mathemat-
44(3):223235, 2012.
[46] A. Hmimid, M. Sayyouri, and Hassan. Image classication using novel set of
Charlier moment invariants.
WSEAS Transactions on Signal Processing, 10:156
167, 2014.
[47] A. Hmimid, M. Sayyouri, and H. Qjidaa. Image classication using a new set
of separable two-dimensional discrete orthogonal invariant moments.
Electronic Imaging,
Journal of
23(1):013026, 2014.
[48] T.V. Hoang and S. Tabbone. Fast generic Polar Harmonic Transforms.
Transactions on Image Processing,
IEEE
23(7):29612971, 2014.
[49] T.V. Hoang and S. Tabbone. Generic polar harmonic transforms for invariant
image representation.
32(8):497509, 2014.
[50] B. Honarvar, R. Paramesran, and C.L. Lim. Image reconstruction from a complete set of geometric and complex moments.
Signal Processing,
98:224232,
2014.
[51] S.J. Hosaini, S. Alirezaee, M. Ahmadi, and S.V.-A.D. Makki.
Comparison of
the Legendre, Zernike and Pseudo-Zernike moments for feature extraction in
[52]
[53]
iris recognition. In International Conference on Computational Intelligence and
Communication Networks (CICN), pages 225228, 2013.
K.M. Hosny. Exact Legendre moment computation for gray level images. Pattern
Recognition, 40(12):35973605, 2007.
K.M. Hosny. Fast computation of accurate Zernike moments. Journal of RealTime Image Processing, 3(1-2):97107, 2008.
[54] K.M. Hosny.
Fast and low-complexity method for exact computation of 3D
24
G.A. Papakostas
Legendre moments.
32(9):13051314, 2011.
[55] K.M. Hosny. Image representation using accurate orthogonal Gegenbauer moments.
[56] K.M. Hosny.
32(6):795804, 2011.
Fast computation of accurate GaussianHermite moments for
image processing applications.
22(3):476485, 2012.
[57] K.M. Hosny, G.A. Papakostas, and D.E. Koulouriotis. Accurate reconstruction
of noisy medical images using orthogonal moments. In
ference on Digital Signal Processing (DSP),
18th International Con-
pages 16, 1-3 July 2013.
[58] H. Hse and A.R. Newton. Sketched symbol recognition using Zernike moments.
In
International Conference on Pattern Recognition (ICPR),
volume 1, pages
367370, 2004.
[59] H.T. Hu, Y.D. Zhang, C. Shao, and Q. Ju. Orthogonal moments based on exponent functions: Exponent-Fourier moments.
2606, 2014.
[60] M.K. Hu. Visual pattern recognition by moment invariants.
on Information Theory,
47(8):2596
IRE Transactions
8(2):179187, 1962.
[61] S.K. Hwang and W.Y. Kim. A novel approach to the fast computation of Zernike
moments.
39(11):20652076, 2006.
[62] M.B. Imamoglu, G. Ulutas, and M. Ulutas. Detection of copy-move forgery using
krawtchouk moment. In
Engineering (ELECO),
International Conference on Electrical and Electronics
pages 311314, 2013.
[63] Z. Iscan, Z. Dokur, and T. Olmez. Tumor detection by using Zernike moments on
segmented magnetic resonance brain images.
37(3):25402549, 2010.
[64] M. Jesus, M. Requena, P. Moscato, and Manuel Ujaldon. Ecient data partitioning for the GPU computation of moment functions.
Distributed Computing,
Journal of Parallel and
74(1):19942004, 2014.
[65] Z. Ji, Q. Chen, Q.S. Sun, and D.S. Xia. A moment-based nonlocal-means algorithm for image denoising.
Information Processing Letters, 109(23-24):1238
1244, 2009.
[66] V.G. Kaburlasos, S.E. Papadakis, and G.A. Papakostas.
Lattice computing
extension of the FAM neural classier for human facial expression recognition.
IEEE Transactions on Neural Networks and Learning Systems,
24(10):1526
1538, 2013.
[67] H.R. Kanan and K. Faez. Ga-based optimal selection of PZMI features for face
recognition.
Applied Mathematics and Computation,
205(2):706715, 2008.
[68] E.G. Karakasis, G.A. Papakostas, and D.E. Koulouriotis.
using quaternion color moments.
Pattern recognition
In D.B. Vincent, editor,
tion: Practices, Perspectives and Challenges,
Pattern Recogni-
chapter 5, pages 153176. Nova
Publishers, 2013.
[69] E.G. Karakasis, G.A. Papakostas, D.E. Koulouriotis, and V.D. Tourassis. Generalized dual Hahn moment invariants.
46(7):19982014,
2013.
[70] E.G. Karakasis, G.A. Papakostas, D.E. Koulouriotis, and V.D.Tourassis. A unied methodology for computing accurate quaternion color moments and moment invariants.
IEEE Transactions on Image Processing, 23(2):596611, Febru-
25
ary 2014.
[71] A. Khotanzad and Y.H. Hong. Invariant image recognition by Zernike moments.
IEEE Transactions on Pattern Analysis and Machine Intelligence, 12(5):489497,
1990.
[72] H.S. Kim and H.K. Lee.
Invariant image watermark using Zernike moments.
IEEE Transactions on Circuits and Systems for Video Technology,
775, 2003.
[73] W.Y. Kim and Y.S. Kim. Robust rotation angle estimator.
on Pattern Analysis and Machine Intelligence,
13(8):766
IEEE Transactions
21(8):768773, 1999.
[74] E.C. Kintner. On the mathematical properties of the Zernike polynomials.
Acta: International Journal of Optics,
Optica
23(8):679680, 1976.
[75] S.M. Lajevardi and Z.M. Hussain. Higher order orthogonal moments for invariant
facial expression recognition.
Digital Signal Processing, 20(6):17711779, 2010.
[76] J. Lee and H. Ko. Gradient-based local ane invariant feature extraction for
mobile robot localization in indoor environments.
29(14):19341940, 2008.
[77] J.L. Li and D.W. Zhang. Camera calibration with a near-parallel imaging system
Opt. Eng.,
based on geometric moments.
50(2):023601, 2011.
[78] L. Li, S. Li, A. Abraham, and J.S. Pan.
Geometrically invariant image wa-
termarking using Polar Harmonic Transforms.
199:119,
2012.
[79] L. Li, S. Li, H. Zhu, and X. Wu.
Detecting copy-move forgery under ane
Computers & Electrical Engineering, Available
online 15 December 2013, ISSN 0045-7906,. in press.
transforms for image forensics.
[80] W. Li, C. Xiao, and Y. Liu.
Low-order auditory Zernike moment:
approach for robust music identication in the compressed domain.
Journal on Advances in Signal Processing,
a novel
EURASIP
2013(1):132, 2013.
[81] Y. Li. Image copy-move forgery detection based on polar cosine transform and
approximate nearest neighbor searching.
Forensic Science International,
224(1-
3):5967, 2013.
[82] S. Liao, A. Chiang, Q. Lu, and M. Pawlak.
via Gegenbauer moments. In
(ICPR),
volume 3, pages 485488, 2002.
[83] S.X. Liao and M. Pawlak.
analysis.
Chinese character recognition
International Conference on Pattern Recognition
On the accuracy of Zernike moments for image
20(12):13581364, 1998.
[84] Z. Liu, Q. Li, Z. Xia, and Q. Wang. Target recognition of ladar range images
using even-order Zernike moments.
Appl. Opt.,
51(31):75297536, 2012.
[85] Z. Liu and H. Wang. A novel speech content authentication algorithm based on
BesselFourier moments.
24:197208, 2014.
[86] Z.H. Liu and H.X. Wang. PseudoZernike moments-based audio content authentication algorithm robust against feature-analysed substitution attack.
timedia Tools Applicattions.
Mul-
in press.
[87] W. Liyun, L. Hefei, Z. Fuhao, L. Zhengding, and W. Zhendi. Spermatogonium
image recognition using Zernike moments.
in Biomedicine,
95(1):1022, 2009.
Computer Methods and Programs
26
G.A. Papakostas
[88] L. Ma, T. Tan, Y. Wang, and D. Zhang. Local intensity variation analysis for
iris recognition.
[89] Babak Mahdian and Stanislav Saic.
37(6):12871298, 2004.
Detection of copy-move forgery using a
method based on blur moment invariants.
Forensic Science International, 171(2-
3):180189, 2007.
[90] J.V. Marcos and G. Cristobal. Texture classication using discrete Tchebichef
moments,.
J. Opt. Soc. Am. A,
30(8):15801591, 2013.
[91] D. Marin, A. Aquino M.E. Gegundez-Arias, and J.M. Bravo.
A new super-
vised method for blood vessel segmentation in retinal images by using gray-level
and moment invariants-based features.
IEEE Transactions on Medical Imaging,
30(1):146158, 2011.
[92] C. Martinez-Ortiz and J. Zunic. Measuring cubeness in the limit cases.
Mathematics and Computation,
Applied
217(21):88608865, 2011.
[93] J. Mennesson, C. Saint-Jean, and L. Mascarilla. Color FourierMellin descriptors
for image recognition.
40:2735, 2014.
[94] F. Mindru, T. Tuytelaars, L. Van Gool, and T. Moons. Moment invariants for
recognition under changing viewpoint and illumination.
Image Understanding,
Computer Vision and
94(1-3):327, 2004.
[95] R. Mukundan. Some computational sspects of discrete orthonormal moments.
13(8):10551059, 2004.
[96] R. Mukundan, S.H. Ong, and P.A. Lee. Image analysis by Tchebichef moments.
10(9):13571364, 2001.
[97] R. Mukundan and K.R. Ramakrishnan.
Zernike moments.
[98] R. Mukundan and K.R. Ramakrishnan.
Theory and Appplications.
Fast computation of Legendre and
28(9):14331442, 1995.
Moment Functions in Image Analysis:
World Scientic, Singapore, 1998.
[99] K. Nakagaki and R. Mukundan. A fast 4x4 forward discrete Tchebichef transform
algorithm.
IEEE Signal Processing Letters,
14(10):684687, 2007.
[100] M. Novotni and R. Klein. 3D Zernike descriptors for content based shape retrieval. In
ACM Symposium on Solid Modeling and Applications (SM),
pages
216225, 2003.
[101] G.A. Papakostas. Improving the recognition performance of moment features
by selection. In U. Stanczyk and L.C. Jain, editors,
and Pattern Recognition.
Feature Selection for Data
Springer-Verlag, in press.
[102] G.A. Papakostas, Y.S. Boutalis, D.A. Karras, and B.G. Mertzios. Fast numerically stable computation of orthogonal FourierMellin moments.
Vision,
IET Computer
1(1):1116, 2007.
[103] G.A. Papakostas, Y.S. Boutalis, D.A. Karras, and B.G. Mertzios. A new class
of Zernike moments for computer vision applications.
177(13):28022819, 2007.
[104] G.A. Papakostas, Y.S. Boutalis, D.A. Karras, and B.G. Mertzios.
Modied
factorial-free direct methods for Zernike and pseudoZernike moment computation.
IEEE Transactions on Instrumentation and Measurement,
58(7):2121
2131, 2009.
[105] G.A. Papakostas, Y.S. Boutalis, D.A. Karras, and B.G. Mertzios. Pattern classication by using improved wavelet compressed Zernike moments.
Applied
27
212(1):162176, 2009.
[106] G.A. Papakostas, Y.S. Boutalis, D.A. Karras, and B.G. Mertzios. Ecient computation of Zernike and PseudoZernike moments for pattern classication applications.
Pattern Recognition and Image Analysis,
20(1):5664, 2010.
[107] G.A. Papakostas, Y.S. Boutalis, and B.G. Mertzios. Evolutionary selection of
Zernike moment sets in image processing. In
Signals and Image Processing (IWSSIP),
International Workshop on Systems,
pages 1011, 2003.
[108] G.A. Papakostas, Y.S. Boutalis, C.N. Papaodysseus, and D.K. Fragoulis. Numerical error analysis in Zernike moments computation.
puting,
Image and Vision Com-
24(9):960969, 2006.
[109] G.A. Papakostas, Y.S. Boutalis, C.N. Papaodysseus, and D.K. Fragoulis. Numerical stability of fast computation algorithms of Zernike moments.
Applied
195(1):326345, 2008.
[110] G.A. Papakostas, V.G. Kaburlasos, and T. Pachidis.
Thermal infrared face
recognition based on lattice computing (LC) techniques. In
Conference on Fuzzy Systems (FUZZ),
IEEE International
pages 16, 7-10 July 2013.
[111] G.A. Papakostas, E.G. Karakasis, and D.E.Koulouriotis. Novel moment invariants for improved classication performance in computer vision applications.
43(1):5868, 2010.
[112] G.A. Papakostas, E.G. Karakasis, and D.E. Koulouriotis. Ecient and accurate
computation of geometric moments on gray-scale images.
41(6):18951904, 2008.
[113] G.A. Papakostas, E.G. Karakasis, and D.E. Koulouriotis. Accurate and speedy
computation of image Legendre moments for computer vision applications.
age and Vision Computing,
[114] G.A. Papakostas,
Im-
28(3):414423, 2010.
E.G. Karakasis,
D.E. Koulouriotis,
and V.D. Tourassis.
Moment-based local binary patterns: a novel local descriptor for invariant pattern recognition applications.
Neurocomputing,
99(1):358371, 2013.
[115] G.A. Papakostas, D.E. Koulouriotis, and E.G. Karakasis. A unied methodology
for the ecient computation of discrete orthogonal image moments.
Sciences,
Information
179(20):36193633, 2009.
[116] G.A. Papakostas, D.E. Koulouriotis, and E.G. Karakasis. Computation strategies
of orthogonal image moments: A comparative study.
Computation,
Applied Mathematics and
216(1):117, 2010.
[117] G.A. Papakostas, D.E. Koulouriotis, and V.D. Tourassis.
Feature extraction
based on wavelet moments and moment invariants in machine vision systems.
In Fabio Solari, editor,
Human-Centric Machine Vision,
pages 3142. InTech,
2012.
[118] G.A. Papakostas, E.D. Tsougenis, and D.E. Koulouriotis. Near optimum local
image watermarking using Krawtchouk moments. In
shop on Imaging Systems and Techniques (IST),
IEEEInternational Work-
number 464-467, 1-2 July
2010.
[119] G.A. Papakostas, E.D. Tsougenis, and D.E. Koulouriotis. Moment-based local
image watermarking via genetic optimization.
putation,
227:222236, 2014.
[120] M. Pawlak.
Applied Mathematics and Com-
Image Analysis by Moments: Reconstruction and Computational
28
G.A. Papakostas
Aspects. Wroclaw University of Technology Press, Wroclaw, 2006. [online] Available: http://www.dbc.wroc.pl/dlibra/doccontent?id=1432&from=&dirids=1.
[121] Z. Ping, H. Ren, J. Zou, Y. Sheng, and W. Bo. Generic orthogonal moments:
JacobiFourier moments for invariant image description.
40(4):12451254, 2007.
[122] A. Prata and W. Rusch.
expansion coecients.
Algorithm for computation of Zernike polynomials
Appl. Opt.,
28(4):749754, 1989.
[123] S.P. Priyal and P.K. Bora. A robust static hand gesture recognition system using
geometry based normalizations and Krawtchouk moments.
46(8):22022219, 2013.
[124] H. Rahmalan, N.A. Abu, and S.L. Wong.
Using Tchebichef moment for fast
Pattern Recognition and Image Analysis,
and ecient image compression.
20(4):505512, 2010.
[125] J. Revaud, G. Lavoue, and A. Baskurt. Improving Zernike moments comparison
for optimal similarity and rotation angle retrieval.
Analysis and Machine Intelligence,
[126] S. Rodtook and S.S. Makhanov.
rotation moments invariants.
IEEE Transactions on Pattern
31(4):627636, 2009.
Numerical experiments on the accuracy of
23(6):577586,
2005.
[127] A. Ruiz and M. Ujaldon.
Exploiting Kepler capabilities on Zernike moments.
Annals of Multicore and GPU Programming,
1(1):2737, 2014.
[128] S.J. Ryu, M. Kirchner, M.J. Lee, and H.K. Lee. Rotation invariant localization
of duplicated image regions based on Zernike moments.
Information Forensics and Security,
IEEE Transactions on
8(8):13551370, 2013.
[129] S.J. Ryu, M.J. Lee, and H.K. Lee. Detection of copy-rotate-move forgery using
Zernike moments. In
Information Hiding,
volume 6387 of
LNCS,
pages 5165,
2010.
Three-Dimensional moment invariants.
IEEE
Transactions on Pattern Analysis and Machine Intelligence, PAMI-2(2):127136,
[130] F A. Sadjadi and E.L. Hall.
1980.
[131] M. Shabanifard, M.G. Shayesteh, and M.A. Akhaee. Forensic detection of image
manipulation using the Zernike moments and pixel-pair histogram.
Processing,
IET Image
7(9):817828, 2013.
[132] Z. Shao, H. Shu, J. Wu, B. Chen, and J.L. Coatrieux.
Quaternion Bessel
Fourier moments and their invariant descriptors for object reconstruction and
recognition.
47(2):603611, 2014.
[133] Y. Sheng and L. Shen. Orthogonal FourierMellin moments for invariant pattern
recognition.
J. Opt. Soc. Am.,
11(6):17481757, 1994.
[134] H. Shu, H. Zhang, B. Chen, P. Haigron, and L. Luo.
Tchebichef moments for binary and grayscale images.
Image Processing,
Fast computation of
IEEE Transactions on
19(12):31713180, December 2010.
[135] J.D. Shutler and M.S. Nixon.
description of moving features.
Zernike velocity moments for sequence-based
24(4):343356,
2006.
[136] C. Singh and Pooja. Improving image retrieval using combined features of Hough
transform and Zernike moments.
Optics and Lasers in Engineering, 49(12):1384
29
1396, 2011.
[137] C. Singh and Pooja. An eective image retrieval using the fusion of global and
local transforms based features.
Optics & Laser Technology,
44(7):22492259,
2012.
[138] C. Singh and Pooja. Local and global features based image retrieval system using
orthogonal radial moments.
Optics and Lasers in Engineering,
50(5):655667,
2012.
[139] C. Singh and S.K. Ranade. A high capacity image adaptive watermarking scheme
with radial harmonic Fourier moments.
23(5):1470
1482, 2013.
[140] C. Singh and S.K. Ranade.
Image adaptive and high-capacity watermarking
system using accurate Zernike moments.
IET Image Processing,
8(7):373382,
2014.
[141] C. Singh and A.M. Sahan. Face recognition using complex wavelet moments.
47:256267, 2013.
[142] C. Singh and E. Walia. Fast and numerically stable methods for the computation
of Zernike moments.
43(7):24972506, 2010.
[143] C. Singh and E. Walia. Algorithms for fast computation of Zernike moments and
their numerical stability.
[144] C. Singh, E. Walia, and Poojaand R. Upneja.
29(4):251259, 2011.
Analysis of algorithms for fast
computation of pseudoZernike moments and their numerical stability.
Signal Processing,
Digital
22(6):10311043, 2012.
[145] I.M. Spiliotis and B.G. Mertzios.
Real-time computation of two-dimensional
moments on binary images using image block representation.
on Image Processing,
IEEE Transactions
7(11):16091615, 1998.
[146] T. Suk and J. Flusser. Combined blur and ane moment invariants and their
use in pattern recognition.
36(12):28952907, 2003.
Computer
Analysis of Images and Patterns, volume 5702 of LNCS, pages 334341. Springer
[147] T. Suk and J. Flusser. Ane moment invariants of color images. In
Berlin Heidelberg, 2009.
[148] T. Suk and J. Flusser. Ane moment invariants generated by graph method.
44(9):20472056, 2011.
[149] T. Suk and J. Flusser. Tensor method for constructing 3D moment invariants. In
[150]
International Conference on Computer Analysis of Images and Patterns (CAIP),
volume 6855 of LNCS, pages 212219, Seville,Spanien, 2011.
M.R. Teague. Image analysis via the general theory of moments. J. Opt. Soc.
Am., 70(8):920930, 1980.
[151] C.H. Teh and R.T. Chin.
On image analysis by the methods of moments.
10(4):496
513, 1988.
[152] K.H. Thung, R. Paramesran, and C.L. Lim. Content-based image quality metric
using similarity measure of moment vectors.
45(6):2193
2204, 2012.
[153] P. Toharia, O.D. Robles, R. Suarez, J.L. Bosque, and L. Pastor. Shot boundary
detection using Zernike moments in multi-GPU multi-CPU architectures.
of Parallel and Distributed Computing,
72(9):11271133, 2012.
Journal
30
G.A. Papakostas
[154] E.D. Tsougenis, G.A. Papakostas, and D.E. Koulouriotis. Image watermarking
via separable moments.
Multimedia Tools and Applications.
in press.
[155] E.D. Tsougenis, G.A. Papakostas, and D.E. Koulouriotis. Introducing the separable moments for image watermarking in a tottaly moment-oriented framework.
In
18th International Conference on Digital Signal Processing (DSP), pages 16,
2013.
[156] E.D. Tsougenis, G.A. Papakostas, D.E. Koulouriotis, and E.G. Karakasis. Adaptive color image watermarking by the use of quaternion image moments.
Systems with Applications,
Expert
41(14):64086418, 2014.
[157] E.D. Tsougenis, G.A. Papakostas, D.E. Koulouriotis, E.G. Karakasis, and D.A.
Karras. Color image watermarking via Quaternion Radial Tchebichef moments.
In
IEEE International Conference on Imaging Systems and Techniques (IST),
pages 101105, 2013.
[158] E.D. Tsougenis, G.A. Papakostas, D.E. Koulouriotis, and V.D. Tourassis. Performance evaluation of moment-based watermarking methods: a review.
of Systems and Software,
Journal
85(8):18641884, 2012.
[159] E.D. Tsougenis, G.A. Papakostas, D.E. Koulouriotis, and V.D. Tourassis. Towards adaptivity of image watermarking in polar harmonic transforms domain.
54:8497, 2013.
[160] M. Tuceryan. Moment-based texture segmentation.
15(7):659668, 1994.
[161] P.K. Tzouveli, K.S. Ntalianis, and S.D. Kollias. Video object watermarking based
on moments. In
LNCS,
Visual Content Processing and Representation, volume 3893 of
pages 6875, 2006.
[162] E. Walia, C. Singh, and A. Goyal.
On the fast computation of orthogonal
FourierMellin moments with improved numerical stability.
Time Image Processing,
Journal of Real-
7(4):247256, 2012.
[163] T.J. Wang and T.W. Sze.
The image moment method for the limited
range CT image reconstruction and pattern recognition.
34(11):21452154, 2001.
[164] W. Wang and J.E. Mottershead. Adaptive moment descriptors for full-eld strain
and displacement measurements.
Design,
Journal of Strain Analysis for Engineering
48(1):1635, 2012.
[165] X.Y. Wang, T.X. Ma, and P.P. Niu.
A pseudo-Zernike moment based audio
watermarking scheme robust against desynchronization attacks.
Electrical Engineering,
Computers &
37(4):425443, 2011.
[166] X.Y. Wang, E.N. Miao, and H.Y. Yang. A new SVM-based image watermarking
using GaussianHermite moments.
Applied Soft Computing,
12(2):887903,
2012.
[167] X.Y. Wang, P.P. Niu, H.Y. Yang, C.P. Wang, and A.l. Wang. A new robust color
image watermarking using local quaternion exponent moments.
Sciences.
Information
in press.
[168] C.Y. Wee and R. Paramesran. On the computational aspects of Zernike moments.
25(6):967980, 2007.
[169] C.Y. Wee, R. Paramesran, R. Mukundan, and X. Jiang. Image quality assessment
by discrete orthogonal moments.
Pattern Recognition, 43(12):40554068, 2010.
[170] C.H. Wei, Y. Li, W.Y. Chau, and C.T. Li.
31
Trademark image retrieval using
synthetic features for describing global shape and interior structure.
Recognition,
Pattern
42(3):386394, 2009.
[171] C.H. Wei, Y. Li, and P.J. Huang. Mammogram retrieval through machine learning within BIRADS standards.
Journal of Biomedical Informatics,
44(4):607
614, 2011.
[172] H. Wu, J.L. Coatrieux, and H. Shu. New algorithm for constructing and computing scale invariants of 3D Tchebichef moments.
Engineering,
Mathematical Problems in
2013:Article ID 813606, 8 pages, 2013.
[173] K. Wu, H. Shu, and J.L. Dillenseger. Region and boundary feature estimation on
ultrasound images using moment invariants.
in Biomedicine,
Computer Methods and Programs
113(2):446455, 2014.
[174] B. Xiao, J.F. Ma, and J.T. Cui. Radial Tchebichef moment invariants for image recognition.
Journal of Visual Communication and Image Representation,
23(2):381386, 2011.
[175] B. Xiao, J.F. Ma, and X. Wang.
Image analysis by BesselFourier moments.
43(8):26202629, 2010.
[176] X. Xiao, P. Wang, and K.C. Chou.
Predicting protein structural classes with
pseudo amino acid composition: An approach using geometric moments of cellular automaton image,.
Journal of Theoretical Biology, 254(3):691696, 2008.
[177] Y. Xin, S. Liao, and M. Pawlak. Circularly orthogonal moments for geometrically
robust image watermarking.
[178] Y. Xin, M. Pawlak, and S. Liao.
in polar coordinates.
40(12):37403752, 2007.
Accurate computation of Zernike moments
2007.
[179] D. Xu and H. Li. Geometric moment invariants.
16(2):581587,
Pattern Recognition, 41(1):240
249, 2008.
[180] D. Xu, Y. Zheng, Y. Gao, and D. Wang.
Parallel computation for discrete
orthogonal moments of images using graphic processing unit.
mation & Computational Science,
Journal of Infor-
9(3):611618, 2012.
[181] H. Yan, Z. Zhenwen, and L. Jianfeng. Human action classication based on the
International Conference on Measuring Technology and
Mechatronics Automation (ICMTMA), pages 305308, 2014.
B. Yang and M. Dai. Image analysis by GaussianHermite moments. Signal
Processing, 91(10):22902303, 2011.
3D Zernike moment.
[182]
[183] B. Yang, G. Li, H. Zhang, and M. Dai.
of GaussianHermite moments.
Rotation and translation invariants
32(9):12831298,
2011.
[184] P.T. Yap, X. Jiang, and A.C. Kot. Two-dimensional Polar Harmonic Transforms
for invariant image representation.
Machine Intelligence,
IEEE Transactions on Pattern Analysis and
32(7):12591270, 2010.
[185] P.T. Yap and R. Paramesran. Local watermarks based on Krawtchouk moments.
In
IEEE Proceedings Region Conference TENCON, volume 2, pages 7376, 21-
24 Nov. 2004.
[186] P.T. Yap, R. Paramesran, and S.H. Ong. Image analysis by Krawtchouk moments.
12(11):13671377, 2003.
32
G.A. Papakostas
[187] P.T. Yap, R. Paramesran, and S.H. Ong. Image analysis using Hahn moments.
29(11):2057
2062, 2007.
[188] J. Yin, A.R. De Pierro, and M. Wei. Analysis for the reconstruction of a noisy
signal based on orthogonal moments.
Applied Mathematics and Computation,
132(2-3):249263, 2002.
[189] W. Yuanbin, Z. Bin, and Y. Tianshun. Projective invariants of co-moments of
2D images.
43(10):32333242, 2010.
[190] H.L. Zhai, Y.Y. Zhai, P.Z. Li, and Y.L. Tian. A simple approach to quantitative
analysis using three-dimensional spectra based on selected Zernike moments.
Analyst,
138(2):683687, 2013.
[191] F. Zhang, S.Q. Liu, D.B. Wang, and W. Guan.
frared image using wavelet moment invariants.
Aircraft recognition in in-
27(4):313318, 2009.
[192] H. Zhang, H. Shu, G. Coatrieux, J. Zhu, Q.M.J. Wu, Y. Zhang, H. Zhu, and
L. Luo. Ane Legendre moment invariants for image watermarking robust to
geometric distortions.
20(8):2189
2199, 2011.
[193] H. Zhang, H. Shu, G.N. Han, G. Coatrieux, and L. Luo anf J.L. Coatrieux.
Blurred image recognition by Legendre moment invariants.
on Image Processing,
IEEE Transactions
19(3):596611, 2010.
[194] H. Zhang, H.Z. Shu, P. Haigron, B.S. Li, and L.M. Luo. Construction of a complete set of orthogonal FourierMellin moment invariants for pattern recognition
applications.
28(1):3844, 2010.
[195] L. Zhang, G. Qian, and W. Xiao. Geometric distortions invariant blind second
generation watermarking technique based on Tchebichef moment of original
image.
Journal of Software,
1(9):22832294, 2007.
[196] H. Zhu. Image representation using separable two-dimensional continuous and
discrete orthogonal moments.
45(4):15401558, 2010.
[197] H. Zhu, M. Liu, H. Ji, and Y. Li. Combined invariants to blur and rotation using
Zernike moment descriptors.
Pattern Analysis and Applications, 13(3):309319,
2010.
[198] H. Zhu, M. Liu, and Y. Li. The RST invariant digital image watermarking using
Radon transforms and complex moments.
Digital Signal Processing, 20(6):1612
1628, 2010.
[199] H. Zhu, H. Shu, J. Liang, L. Luo, and J.L. Coatrieux. Image analysis by discrete
orthogonal Racah moments.
Signal Processing,
87(4):687708, 2007.
[200] H. Zhu, H. Shu, T. Xia, L. Luo, and J.L. Coatrieux.
invariants of Tchebichef moments.
Translation and scale
Pattern Recognition, 40(9):25302542, 2007.
[201] H. Zhu, H. Shu, J. Zhou, L. Luo, and J.L. Coatrieux. Image analysis by discrete
orthogonal dual Hahn moments.
28(13):1688
1704, 2007.
[202] D. Zunic and J. Zunic. Shape ellipticity from Hu moment invariants.
[203] J. Zunic and P.L. Rosin.
Applied
226:406414, 2014.
Measuring linearity of open planar curve segments.
29(12):873879, 2011.

vol01ch01 - Over 50 Years of Image Moments and Moment Invariants

Transcription

Similar documents

TROY, NY MAKEUP EXAM NO. 2 INTRODUCTION TO ENGINEERING ANALYSIS

view and our Brochure

Album Notes

June 2015 - Head Start of Lane County

Scam 2

TapeMyDay Selfie flyer

4 New Moments Every Marketer Should Know